Early warning models in real time

Using a unique set of banking data containing both originally reported and subsequently revised financial variables, we find adverse revisions to accounting statements are associated with downgrades in supervisory ratings. To assess the financial significance of the revisions, we compare the ability of the original and revised data to map into exam ratings. The relationship between accounting data and exam results is significantly stronger for revised data than for real-time data. Our findings document significant differences between real-time and revised banking data, highlight the auditing role of bank exams, and provide a more realistic assessment of early warning model accuracy.